Seismicity at the convergent plate boundary offshore Crete, Greece, observed by an amphibian network

(1)

HAL Id: hal-00535497

https://hal.archives-ouvertes.fr/hal-00535497

Submitted on 11 Nov 2010

HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.

L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.

Seismicity at the convergent plate boundary offshore Crete, Greece, observed by an amphibian network

D. Becker, T. Meier, M. Bohnhoff, H.-P. Harjes

To cite this version:

D. Becker, T. Meier, M. Bohnhoff, H.-P. Harjes. Seismicity at the convergent plate boundary offshore

Crete, Greece, observed by an amphibian network. Journal of Seismology, Springer Verlag, 2009, 14

(2), pp.369-392. �10.1007/s10950-009-9170-2�. �hal-00535497�

(2)

DOI 10.1007/s10950-009-9170-2

ORIGINAL ARTICLE

Seismicity at the convergent plate boundary offshore Crete, Greece, observed by an amphibian network

D. Becker·T. Meier·M. Bohnhoff·H.-P. Harjes

Received: 1 July 2008 / Accepted: 28 April 2009 / Published online: 14 May 2009

Abstract We investigate microseismic activity at

the convergent plate boundary of the Hellenic subduction zone on- and offshore south-eastern Crete with unprecedented precision using record- ings from an amphibian seismic network. The net- work configuration consisted of up to eight ocean bottom seismometers as well as five temporary short-period and six permanent broadband sta- tions on Crete and surrounding islands. More than 2,500 local and regional events with magnitudes up to M

L=4.5

were recorded during the time period July 2003–June 2004. The magnitude of completeness varies between 1.5 on Crete and adjacent areas and increases to 2.5 in the vicinity of the Strabo trench 100 km south of Crete. Tests with different localization schemes and velocity models showed that the best results were obtained from a probabilistic earthquake localization using a 1-D velocity model and corresponding station

D. Becker (

B

⁾

Institute of Geophysics, Hamburg University, Bundesstr. 55, 20146 Hamburg, Germany e-mail: dirk.becker@zmaw.de

T. Meier·H.-P. Harjes

Institute of Geology, Mineralogy and Geophysics, Ruhr University Bochum, Bochum, Germany M. Bohnhoff

GeoForschungsZentrum Potsdam, Potsdam, Germany

corrections obtained by simultaneous inversion.

Most of the seismic activity is located offshore of central and eastern Crete and interpreted to be associated with the intracrustal graben sys- tem (Ptolemy and Pliny trenches). Furthermore, a significant portion of events represents interplate seismicity along the NNE-ward dipping plate in- terface. The concentration of seismicity along the Ptolemy and Pliny trenches extends from shal- low depths down to the plate interface and in- dicates active movement. We propose that both trenches form transtensional structures within the Aegean plate. The Aegean continental crust be- tween these two trenches is interpreted as a fore- arc sliver as it exhibits only low microseismic activity during the observation period and little or no internal deformation. Interplate seismicity between the Aegean and African plates forms a 100-km wide zone along dip from the Strabo trench in the south to the southern shore-line of Crete in the north. The seismicity at the plate contact is randomly distributed and no indications for locked zones were observed. The plate contact below and north of Crete shows no microseismic activity and seems to be decoupled. The crustal seismicity of the Aegean plate in this area is gen- erally confined to the upper 20 km in agreement with the idea of a ductile deformation of the lower crust caused by a rapid return flow of meta- morphic rocks that spread out below the forearc.

In the region of the Messara half-graben at the

(3)

south coast of central Crete, a southward dipping seismogenic structure is found that coalesces with the seismicity of the Ptolemy trench at a depth of about 20 km. The accretionary prism south of Crete indicated by the Mediterranean Ridge showed no seismic activity during the observation period and seems to be deforming aseismically.

Keywords Hellenic subduction zone·

Microseismicity

·

Forearc sliver

·

Crete

·

Amphibian seismic network

·

Return flow

1 Introduction

1.1 Tectonic setting of the Hellenic subduction zone

The Hellenic subduction zone, referred to as HSZ hereafter, is the seismically most active region in Europe. This high seismic activity is caused by the subduction of the oceanic African lithosphere beneath the continental Anatolian–

Aegean lithosphere as manifested by Benioff zone seismicity (Papazachos and Comninakis

1971;

Makropoulos and Burton

1984; Papazachos et al.

2000a, b) and seen in the images of seismic

tomography (Spakman et al.

1993; Papazachos

and Nolet

1997). According to Thomson et al.

(1998), the current active margin south of Crete was initiated approximately 15 Ma ago follow- ing earlier subduction of oceanic basins north of the present day active margin and accretion of continental terranes to Eurasia (Dercourt et al.

1986; Gealey 1988). Neotectonic reconstructions

indicate a migration of the active margin ap- proximately 400–500 km to the south-west during the last 15 Ma (ten Veen and Meijer

1998; ten

Veen and Kleinspehn

2003). The retreat of the

active margin is caused by slab-pull of the down- going African lithosphere (Angelier et al.

1982;

Le Pichon et al.

1995; Laigle et al. 2004). In the

same time interval, the African plate approached Eurasia by about 150 km (Dercourt et al.

1986;

Gealey

1988; Facenna et al. 2003). At present,

the Aegean plate moves towards the south-west at 3.0–3.5 cm/year with respect to stable Eurasia while the African plate exhibits a northward drift

of less than 1 cm/year within the same reference frame (McClusky et al.

2000).

The HSZ in the region of western Crete seems to enter the phase of continent–continent collision (Armijo et al.

1992; ten Veen and Kleinspehn 2003) with the passive continental margin of

Africa entering the subduction (Meier et al.

2004a). While subduction below western Crete

occurs almost normal to the continental margin of the Aegean plate, subduction becomes more and more oblique towards the east (Le Pichon et al.

1995). Offshore central and eastern Crete, the on-

going slab rollback in the Cretan–Rhodes section of the Aegean forearc is the likely cause for the transtensional structures within the continental Aegean lithosphere which are the dominating fea- tures south of central and eastern Crete (ten Veen and Kleinspehn

2003). These structures are indi-

cated in Fig.

1

by solid lines. In the vicinity of the Ptolemy and Pliny trenches, wedge-shaped sedi- mentary basins were imaged from wide-aperture seismic profiles (Bohnhoff et al.

2001) and swath-

mapping and seismic reflection profiles in the re- gion of the Pliny and Strabo trenches show en echelon troughs and normal faulting (e.g., Huchon et al.

1982; LePichon et al. 1982; Huguen et al.

2001). Towards the west, this graben system con-

tinues in the Ionian trench which is interpreted as an extensional structure within the Aegean lithosphere (Lallemant et al.

1994) that shows

currently little microseismic activity (Meier et al.

2004b). All these so-called trenches are inter-

preted as structures within the continental Aegean crust while the margin of the Aegean lithosphere is found further south buried below a thick sedi- mentary cover (Lallemant et al.

1994; Mascle et al.

1999; Huguen et al.2001).

1.2 Seismicity of the HSZ

For the area of the HSZ, an extensive historic

catalog (Papazachos et al.

2000a,b), spanning the

last 2,500 years, as well as a number of microseis-

micity studies investigating the entire (Hatzfeld

et al.

1993) or parts of the forearc (e.g., Delibasis

et al.

1999; Becker2000; Meier et al.2004b,2007),

is available in addition to global catalogs covering

the second half of the twentieth century (e.g.,

Engdahl et al.

1998).

(4)

34 35 36

Major Events Depth [km]

<60 >60

Harvard CMT Mw = 4.7 Mw = 5.7 Mw = 6.7

Depth [km]

0 − 50

> 50 Mw = 6 Mw = 7 Mw = 8

Mud Volcanism

Olimpi

United Nations

6 mm/a

35 mm/a

IONIAN

STRABO PLINY

PTOLEMY

o o o o o

o o o

Fig. 1 Historic seismicity and Harvard CMT solutions calculated for the convergent plate boundary of the Hellenic subduction zone. Major seismicity (black and white triangles) with magnitudes Mw≥6.0spans the time period 550 BC–1999 AD (Papazachos et al. 2000a, b) while beach balls represent fault plane solutions for events between 1977 and 2008 (www.globalcmt.org). Solid black lines indicate the locations of the Ionian, Ptolemy, Pliny, and Strabo trenches as seen in the bathymetry and arrows indicate sense of displacement of the transtensional struc-

tures (e.g., Kahle et al. 2000; ten Veen and Kleinspehn 2003). Broken black line indicates the presumed crustal ocean–continent transition. Mud volcano fields in the accretionary prism south of the Pliny and Strabo trenches are indicated by broken magenta lines (Galindo-Zaldivar et al.

1996; Cifci et al.1997). Plate velocities and directions of movement of the African and Aegean plate are indicated by arrows. Velocities are relative to a stable Eurasian reference frame. See McClusky et al. (2000) for details

Interplate seismicity in the HSZ follows the amphitheatrical shape of the Benioff zone from the Kefalonia fault in the west to the Turkish coast north-east of Rhodes in the east. Figure

2

depicts this activity in the vicinity of Crete. Towards the north, the Benioff zone seismicity of the HSZ can be traced to depths of up to 180 km in the eastern part close to Rhodes and depths of approximately 100 km below Milos in the West using the relo- cated events of the ISC catalog (Engdahl et al.

1998; Fig. 2) with much stronger activity in the

central and eastern part than in the west.

According to the catalog of historic earth- quakes of the Aegean region compiled by Papazachos et al. (2000a,

b), the largest docu-

mented event with a magnitude of about M

w

8.3 (Papazachos et al.

2000a,b; Stiros2001) occurred

in 365 AD in the western forearc of the subduction

zone south-west of Crete, and it is suggested that

this event caused the observed coastline uplifts

of up to 9 m on western Crete and up to 2.5 m

on the adjacent island of Antikythira (Pirazzoli

1996; Stiros 2001; Shaw et al. 2008). This event

presumably ruptured the plate interface between

the African and Aegean plates (Papazachos and

Papazachou

1997; Papazachos et al. 1999) on a

lateral extent of 300 km from south of central

Crete up to 22° E (Papazachos

1996). However,

the uplift could as well be explained by an event

within the upper plate below Crete, a slow uplift

(5)

34 35 36 37

A A’

0 100 200 300

Days after 18/07/2003

LIB1 LIB2

LIB3 LIB6 LIB4

LI12 LI14

LI15 LI16

LI17 LI18

LI19

LI10 ZKR SANT

APE

APEZ IDI

KRIS

GVD

KARP

ISC data

Depth [km] Magnitude 0−20

20−40 40−60 60−100

>100

M_w = 3 Mw = 4 M_w = 5 Mw = 6

0 50 100 150 200 250 300

0 20 40 60

Depth [km]

Position on Profile [km]

Ptolemy Pliny

Strabo

0 50 100 150 200 250 300

0 20 40 60

Depth [km]

Ptolemy Pliny

Strabo A’

A

o o o o o

o o o o

Fig. 2 Relocated seismicity according to the ISC catalog during the time period 1963–1999 (Engdahl et al. 1998) for the region around Crete (circles) and microseismic- ity located by temporary onshore networks (Meier et al.

2004a; Becker2000; blue dots). The configuration of the amphibian seismic network evaluated in this study is super- imposed. Triangles mark the OBS locations of LIBNET, diamonds mark temporary short period onshore stations and inverted triangles mark permanent broadband sta- tions from the GEOFON, NOA, and MEDNET networks.

Color-coding of the symbol outline indicates the begin- ning of recording while the color in the symbols interior indicates the end of recording for the respective station.

Corresponding color scale is given in the lower part of the figure. Inset: Hypocenters of ISC catalog events (dark gray dots; Engdahl et al. 1998) and microseismicity recorded with a network on eastern Crete (light gray dots; Becker 2000) projected on a plane along profile AA’. Events with a distance of up to 20 km from the profile are used for the projection

over a longer time period or a series of small events (Shaw et al.

2008). The eastern portion of

the plate interface in this region currently exhibits strong seismicity (Fig.

2).

In instrumental seismicity data (Engdahl et al.

1998), crustal seismicity within the upper plate is

found in the central and eastern offshore region south of Crete in the vicinity of the Ptolemy trench and further south between the Pliny and Strabo trench. Some historic events are located in the same area (Papazachos et al.

2000a,b) although it

is generally not possible to discriminate between intra- and interplate activity for these events due to weakly constrained hypocentral depths. Re- markably, almost all of the shallow strong historic activity with magnitudes of M

_w≥6

locates close

to the Cretan shore in the area of the Ptolemy trench or even onshore while the medium sized instrumental activity which is generally in the range of M

_w

3.5–5 is found in its majority further south between Pliny and Strabo. However, we cannot exclude that this difference is caused by poor hypocenter precision of the historic events.

Most of this shallow activity is likely to be caused

by relative movements along transtensional struc-

tures caused by the oblique subduction and

slab rollback (Kahle et al.

2000; ten Veen and

Kleinspehn

2003; Bohnhoff et al. 2005). In the

western offshore area, seismicity is generally con-

fined to the plate interface with low seismic

activity at the Ionian trench. Historic and instru-

mental data are in good agreement for this region.

(6)

Seismic activity on Crete itself is generally lower and occurs predominantly on north–south striking normal faults in the west that are capable of pro- ducing events up to M

_w

5.5 (de Chabalier et al.

1992). Furthermore, a trend of increasing activity

towards the east with some strong historic events along the south coast of central and eastern Crete is observed. Towards the north in the Cretan Sea, hardly any shallow seismic event is observed and most of the seismicity occurs at intermediate depth.

Microseismicity studies in the forearc of the HSZ complement the historic and instrumental data sets. Several microseismic studies were con- ducted on the islands of Crete and Gavdos with temporary short period networks to study the distribution of microseismic activity in the region.

In 1988, a study of microseismic activity covering the whole southern Aegean Sea recorded only sparse microseismic activity onshore Crete while most activity occurred south of Crete between the Pliny trench and the coast (Hatzfeld et al.

1993). A study conducted in 1995 in central Crete

(Delibasis et al.

1999), however, found significant

microseismic activity in the Messara graben, the Heraklion Basin, and in the southern offshore re- gion at the Ptolemy trench. This latter activity was also observed during a 2000/2001 field campaign aimed at studying the microseismic activity in cen- tral Crete (Meier et al.

2004a). Field campaigns in

the years 1996–2002 mapped microseismic activity at the plate interface (Meier et al.

2004a) and

identified an offset between the southern border of the Aegean lithosphere and the southern bor- der of active interplate seismicity. Most of the offshore activity in the western part originated at the plate contact with only little activity in the Ionian trench. A microseismic study in eastern Crete in early 1999 found abundant activity on- shore eastern Crete as well as in the southern offshore region (Becker

2000).

The only ocean bottom seismometers (OBS) network operated so far in the Libyan Sea con- sisted of five analog OBS stations that were de- ployed southeast of Crete for 8 days in July 1988.

One thousand microearthquakes were detected of which more than 140 were located (Kovachev et al.

1991, 1992). The authors interpreted the

spatial distribution of the hypocenters as an en-

echelon positioning of the subduction zone man- ifested bathymetrically in the Strabo and Pliny trenches.

The aim of this study is to better resolve the seismogenic regions in the area offshore central and eastern Crete. The study region includes the offshore graben system as well as the plate contact in that area and the accretionary prism to the south. The area of investigation had not yet been analyzed for (micro)seismic activity using modern digital ocean bottom instrumentation combined with stations on Crete, allowing to decrease the magnitude detection threshold to below M3.5.

The inset of Fig.

2

shows a cross-section incor- porating information from the global ISC cata- log and data from an onshore network (Becker

2000). While the global data are too sparse to

accurately define the seismogenic structures, the local network is restricted to events onshore and close to the Cretan shore. The installation of an amphibian network covering the entire trench sys- tem south of central and eastern Crete on the other hand should allow a precise localization of the seismogenic zones in this area by significantly improving the location accuracy of offshore events and reducing the detection threshold.

Furthermore, we aim at investigating the influ- ence of different velocity models and localization techniques on the hypocentral distribution and location precision. Therefore, two different 1-D and a simple 3-D model are tested in conjunc- tion with a gradient method and a probabilistic localization scheme for the determination of the hypocentral coordinates. The performance of the different localization methods is compared and the best constrained seismicity distribution is used to interpret the results with regard to the regional seismotectonic setting.

2 Experimental setup and data processing

The initial station distribution of the amphibian network (LIBNET) aimed at monitoring the en- tire trench system SE of Crete as well as the supposed southern border of the Aegean plate.

Therefore, one OBS (LIB1 in Fig.

2) was placed

at a location presumably on top of the oceanic

African lithosphere as suggested by refraction

(7)

seismic studies (Brönner

2003). With this station,

layout a lower detection threshold, a better lo- cation precision for crustal seismicity within the Aegean plate as well as better depth constraints for events at the plate boundary were expected compared to land-based networks or networks of regional or global scale. Thus, the identification of active structures within the continental Aegean crust south of Crete as well as the determina- tion of the updip limit of interplate seismicity becomes a realistic target. Furthermore, it permits a glimpse at the accretionary complex to the south with its mud volcanism (Fig.

1; Kopf2002).

The LIBNET campaign was carried out in five separate observation phases during which up to eight OBS stations were deployed for a record- ing period of up to 60 days. After each de- ployment period, the OBS were recovered and then redeployed. Except for one OBS in the last observation phase which was equipped with a CME-40 broadband sensor, all OBS stations were equipped with three short-period SM-6 B-coil geophones (one vertical and two horizontal com- ponents) with a natural frequency of 4.5 Hz. Data loggers were of the SEDIS type (www.geopro.

com/instrumentation/sedis-v.html). The sampling

rate was 128 or 250 Hz, respectively, and a clock drift correction was applied to the data after sta- tion recovery. A maximum of six stations pro- duced reliable recordings within each recording period.

The onshore part of LIBNET consisted of five short-period stations equipped with Mark L4-3D seismometers and PDAS-100 data loggers oper- ated at 50 Hz sampling frequency. In addition, permanent broadband stations were incorporated in the data analysis equipped with either STS-2 Streckeisen seismometers or Lennartz LE-20/3D sensors. These stations are part of the GEOFON (Hanka and Kind

1994), the National Observa-

tory of Athens (NOA) and MEDNET networks, respectively. Data from these stations were pro- vided by the GEOFON data center in Potsdam, Germany (http://www.gfz-potsdam.de/geofon/).

Due to the inferior data quality of the southern- most OBS stations (Fig.

3a), presumably caused

by the thick sedimentary cover and week coupling conditions, the last two deployment periods were carried out with a modified network geometry.

Station coverage in the northern part of the study region was densified by deploying all OBS stations between the south coast of Crete and the Strabo trench (Fig.

2). This allowed to obtain higher data

recovery rates for the last two deployment periods and to further lower the detection threshold for events in the seismically active offshore region between Crete and the Strabo trench.

During data processing, continuous waveforms from LIBNET stations and the temporary short period onshore stations were filtered between 2 and 20 Hz, and a STA/LTA trigger (Robinson

1967) was applied for event detection. The sub-

sequent application of a coincidence trigger re- quired a minimum of three stations triggering within 19 s. This time interval was chosen due to the interstation distances and the assumption of low seismic velocities within the thick sedimentary cover.

High seismic activity was observed throughout the whole observation campaign with occasional bursts of activity consisting of more than 100 events per day. However, these events were of- ten small and only recorded at one or two sta- tions and thus not included in the further data processing. The best data quality for OBS stations is found in the region between Crete and the Pliny trench while stations further south exhibited considerably lower signal to noise ratios (SNRs) and often very long signal coda (Fig.

3a). This

variation in data quality might be explained by a thick, water-saturated sediment cover present in the southern part which also prevented sig- nal penetration during active seismic experiments in the region (Brönner

2003). The few intraslab

events detected from below Crete and the Sea of Crete, on the other hand, exhibit high SNRs on the northernmost OBS stations (Fig.

3b) and

indicate the waveguide quality of the subducting African slab.

The triggering procedure led to the identifica-

tion of 3,257 potential events covering the time

period July 2003–June 2004. After visual inspec-

tion and elimination of teleseismic and multiple

triggered events, a total of 2,686 local and regional

events were further analyzed. There were 17,173

P- and 15,995 S-phase onsets picked manually on

the vertical component for the P-phase and on

the horizontal components for the S-phase. Local

(8)

0 20 40 60 80 100 120 Time [s]

CCA

CCD

LIB1

LIB2

LIB3

LIB6

10 30 50 70 90 110 130

Time [s]

APE

SANT

CCE

IDI

LI14

LI15

b) a)

Fig. 3 a Z-component waveform examples from a south- ern offshore event located at 34.33° N, 26.05° E at a depth of 20.5 km (depicted by a red star in Fig.2). Data passband filtered between 5 and 20 Hz. Note the marked difference in signal quality between different OBS stations and the long signal coda at stations LIB1 and LIB3 which is typical for many OBS stations in the south of the study region positioned on top of the thick sedimentary cover

of the accretionary complex. Red lines indicate the picked P-onsets. b Z-component waveform examples from an event within the subducting African slab north of Crete 36.70° N, 26.85° E at a depth of 150 km (indicated by a red star in Fig.2). Data passband filtered between 5 and 20 Hz.

Note the high signal to noise ratio for the OBS stations south of Crete indicative of the waveguide qualities of the subducting African slab

Richter magnitudes were calculated for the events from restituted displacement proportional records simulating the recording of a Wood-Anderson seismometer and taking into account a distance correction suggested for events in the Aegean (Kiratzi and Papazachos

1984).

3 Event localization procedure

To achieve an optimal hypocenter determination and to investigate the influence of the localization scheme and the velocity model on the seismic- ity distribution, four different approaches were tested: (1) an initial velocity model derived in an earlier study (Meier et al.

2004a) in conjunction

with a gradient method, (2) a minimum 1-D ve- locity model gained by simultaneous inversion for the velocity model and station corrections, (3) a simple 3-D velocity model used in conjunction with a probabilistic event localization technique, and (4) a probabilistic event localization applied to the minimum 1-D velocity model with corre- sponding station corrections. With these tests, the

influence of the slab and the horizontally varying velocity structure on the seismicity distribution can be investigated. Furthermore, by the proba- bilistic approach, the model space is more com- pletely sampled and no termination of the location procedure in local minima close to the initial event location occurs.

3.1 Initial event localization

Initial event locations were obtained using the hypoinverse2000 location software (Klein

2002)

and a velocity model obtained during earlier seismological and seismic studies in the region (Bohnhoff et al.

2001; Brönner2003; Meier et al.

2004a; Fig. 4a). This velocity model is very sim-

ilar to the one derived by Meier et al. (2004a)

in a microseismicity study for the region of cen-

tral Crete which was obtained by solving for a

minimum 1-D model using the VELEST inver-

sion software (Kissling et al.

1994). A vp/vs

ve-

locity ratio of 1.78 was adopted after evaluation

of Wadati plots for events within the network.

(9)

2 4 6 8 0

20

40

60

80

100

Initial velocity model

Velocity [km/s]

Depth [km]

vp

vs

2 4 6 8

0

20

40

60

80

100

Minimum 1−D model

Velocity [km/s]

Depth [km]

vp

vs

Fig. 4 a Initial velocity model after Meier et al. (2004a), modified according to information from active seismics (Bohnhoff et al. 2001; Brönner 2003). vp/vs ratio is set to 1.78. b Final minimum 1-D velocity model obtained by

VELEST inversion. The large thickness of the first layer is due to the fact that the deepest OBS was 3.6 km below sea level and had to be contained within this layer.vp/vsratio is set to 1.78

Because hypoinverse2000 assumes all stations to be at zero elevation, station delays were estimated for OBS as well as onshore stations according to their respective elevation above or below the top of the velocity model. The starting depth was varied between 5 and 150 km and the hypocenter calculation was terminated after 40 iterations. The solution with the smallest root mean square (rms) value and location uncertainty was adopted as hypocenter. The step size of the starting depth variation was 5 km down to 40 km depth and in- creasingly larger below. This approach was chosen to more completely sample the depth space and avoid termination of the localization program in a local minimum. This significantly increased the number of events converging to a low rms solution within the given number of iterations when com- pared to a single location run with just one fixed starting depth.

3.2 Joint determination of hypocenters and 1-D velocity model

To obtain a more realistic 1-D velocity model for the entire study region and station corrections that account for site-specific velocity anomalies, a si-

multaneous inversion using the VELEST software was performed.

The inversion scheme simultaneously solves for velocity model and station corrections by a least squares formulation minimizing the weighted mis- fit between predicted and observed arrival times.

Details and limitations of this inversion procedure were given by Kissling et al. (1994). This results in a minimum 1-D velocity model with correspond- ing station corrections and hypocenter locations.

Inversion for the minimum 1-D velocity model was performed using version 3.3 of the VELEST software. As input, 435 events from the initial hypoinverse2000 localization (see Fig.

5d) with at

least 6 P readings, an azimuthal gap less than 180°, an rms value of less than 0.7 s and horizontal and vertical errors less than 10 and 15 km, respectively, as indicated by the hypoinverse2000 routine, were used. Onshore station CCD served as reference station. This station exhibits an almost continuous data record for the whole observation period and high data quality. More than 100 starting models which sample a wide range of the model space and contain no low velocity zones for stability reasons were investigated and are depicted in Fig.

5a.

During the inversion, hypocenter and station files

were updated after each VELEST run and, after

(10)

Fig. 5 Results from the VELEST 1-D velocity inversion. a The 10vp

models with the lowest rms misfit (blue)

superposed on all starting models. b Resultant P-velocity models color-coded according to their rms misfits. cvp/vs

ratios obtained by solving for independent

S-velocity models. d Plot of station corrections obtained for the minimum 1-D model.

Also plotted are the epicenters of the 435 events used for the initial inversion run

VP [km/s] V

P [km/s]

Depth [km]

4 6 8

0

10

20

30

40

2 10

50

2 4 6 8 10

0

10

20

30

40

50

Depth [km]

0.12 0.13 0.14 0.15 0.16 0.17 0.18 0.19

rms

24 24.5 25 25.5 26 26.5 27 33.5

34 34.5 35 35.5

+2 sec +1 sec

a)

d)

0 1 2 3 4 5

0

10

20

30

40

50

VP / V S

Depth [km]

0.24 0.25 0.26 0.27

rms

b)

c)

five VELEST runs, all available events were re- located and those satisfying the above mentioned criteria were reselected for the next iteration of the inversion. The number of the final selected events for the different inversion runs covers the range from 301 up to 513 selected events with 85 of the velocity models being constrained by more than 450 events. This reflects the fact that most of the models obtained by inversion are slower than the initial model used for localization.

Models constrained by only slightly more than 300 events have very high velocities which results in the exclusion of many events locating now out- side the network and having an azimuthal gap of more than 180°. These models also have very high average rms values and thus show a worse fit to the data than the models with lower velocities which are constrained by more events. Figure

5a, b

shows the resultant velocity models color-coded according to the model misfit. The velocity model is best constrained between 15 and 35 km depth where the P-velocity is of the order of 6 km/s.

Due to the distribution of the recording stations and the hypocenters, the ray density in this depth

range is very high which allows a good determina- tion of the layer velocities. This can be seen from Fig.

5a were the ten velocity models with the best

rms fit are plotted above the initial velocity models and only exhibit a small scatter in this depth range.

For depths below 35 km, the event density starts

to decrease, resulting in less ray paths constraining

the model because most recorded onsets are direct

waves and not refractions. However, a gradual

stepwise increase of the velocity is observed for

nearly all models for depths below 35 km reaching

values of 7.5–7.9 km/s at 50-km depth for models

with the smallest rms misfit (Fig.

5a). At shallower

depths, the velocity models with the smallest rms

misfits indicate P-velocities of

≤4.5

km/s. The

lack of convergence for velocity models within

the uppermost 10 km might be explained with

the strong lateral velocity variations in that depth

range. Because there was no single preferable ve-

locity model, the arithmetic mean of the ten mod-

els with the lowest rms was adopted as minimum

1-D velocity model (Fig.

4b). Station corrections

for this minimum 1-D velocity model were ob-

tained by a final VELEST inversion run. The

(11)

50 100 150 200 250 0

20 40 60 80

Profile km

Depth [km]

1.5 2.86 4.22 5.58 6.94 8.3

v [km/s]_p

Fig. 6 Cross-section through 3-D velocity model used for the probabilistic earthquake localization with the NonLinLoc code (Lomax et al. 2000). Location of the profile is that of profile C–D in Fig.7

initial 435 events used in the inversion were taken as input for this run. A constant

vp/vs

ratio of 1.78 was assumed and the P-velocities were fixed to those of the minimum 1-D velocity model

by overdamping. Calculated station corrections (Fig.

5d) show increasingly positive values to-

wards the south-east while values for stations close to the south coast of Crete are comparable to onshore stations. Final earthquake locations were computed by running VELEST in “single event mode”.

An attempt to solve for independent S-velocity models was done using

vp

models with an rms mis- fit not larger than 0.13 s and a constant

vp/vs

ratio of 1.78 for starting

vs

models. Because the ten best

vp

models show very similar velocities in the depth range between 15 and 35 km (Fig.

5a), we chose all

models with an rms misfit

≤0.13

s, i.e., the best 52 models, to better sample the velocity space.

Results from this inversion attempt show that in the best constrained depth range between 15 and 35 km the

vp/vs

ratio is almost constant around 1.78–1.80 (Fig.

5c). Towards shallower depths high vp/vs

ratios with values above 2 and sometimes as

Fig. 7 Mapview of the event locations obtained by using the minimum 1-D velocity model and corresponding station corrections obtained by the VELEST inversion in conjunction with the probabilistic NLLoc Oct-tree localization scheme

34 35 36

−2000 2000

2000 2000

2000

3000

3000 3000 3000

3000 3000

3000

34 35 36

A B

C D

OBS PDAS BB−Station Depth [km]

0 − 10 10 − 20 20 − 30 30 − 40

> 40

Magnitude Ml = 1 Ml = 2 Ml = 3 Ml = 4 Ml = 5

o o o o

o o o

(12)

high as 3.5 may indicate the presence of the slow, possibly water-saturated sediments in the south.

For depths compatible with the African mantle lithosphere, rather low

vp/vs

ratios of less than 1.7 are found. However, the results from the

vs

model inversion are not very well constrained and thus are not considered in the further discussion.

3.3 Probabilistic event localization

In order to investigate the influence of the slab and the thick sedimentary cover in the southern part of the network on the seismicity distribution, a simple 3-D velocity model based on a priori ve- locity information from refraction seismic studies (Bohnhoff et al.

2001; Brönner 2003) as well as

seismological receiver function and surface wave dispersion studies (Endrun et al.

2004) was gener-

ated. The velocity model, which consists of cubic cells with side lengths of 1 km, has a dimension of 360

×

275

×

91 km in longitude, latitude, and depth, respectively, and starts 1 km above sea level to contain all recording stations within its dimensions. Islands are modeled as blocks being superimposed on a background velocity model.

The horizontal curvature of the subduction zone is neglected in this simple approach. This velocity model (Fig.

6) also includes the effect of the thick

sedimentary cover in the south by incorporating low P-velocities and large

vp/vs

ratios indicated by the above discussed inversion approach and a gently dipping African slab. This model was used in conjunction with the probabilistic NonLinLoc earthquake localization program (Lomax et al.

2000), using the Oct-tree search mode, to solve

for earthquake hypocenters. This procedure starts with a global sampling of the search space us- ing a coarse grid and then finds the maximum likelihood hypocenter by mapping the probability density function of the earthquake location by an increasingly finer sampling of the search space.

This approach samples a wider range of the solu- tion space than the gradient methods and allows an estimation of the location uncertainties. Station corrections for this model were calculated as the mean travel time residual at the respective station and then applied in the final event localization run.

The area in which seismicity can be located with this scheme is truncated by the edges of the veloc- ity model which are found roughly at 33.25 and 35.68

^◦

N and 23.25 and 27.2

^◦

E. These borders are not exactly parallel to meridians because a rectan- gular grid with constant grid spacing is transferred to geographical coordinates. Earthquakes outside these boundaries are projected onto the edges of the velocity model with large rms values.

Finally, the probabilistic NonLinLoc routine was initialized with the minimum 1-D veloc- ity model and corresponding station corrections and the Oct-tree search algorithm was used to calculate event hypocenters. A mapview of the resulting event distribution for this localization approach can be found in Fig.

7.

3.4 Model performances and localization uncertainties

The general microseismicity distribution remains the same for all four localization approaches (Figs.

8

and

9). This indicates the stability of

the localization results and that the general event distribution is not very sensitive with respect to the choice of different, reasonable velocity models and the localization approach. The influ- ence of the slab on the seismicity distribution is small which can be expected because hardly any intermediate depth activity is observed and most of the seismicity is found either within the crust or at the plate contact and is thus not very sen- sitive to the geometry of the subducting African lithosphere. This can be seen when comparing the results for the 3-D velocity model (Fig.

9c) with

the minimum 1-D velocity model (Fig.

9d) for the

area north of the Pliny trench. Locations in this area are very similar for the 1-D and 3-D case.

Towards the south, however, the influence of

the low velocities and the high

vp/vs

ratios of

the sedimentary layers incorporated in the 3-D

velocity model are clearly evident by the much

shallower depths of the hypocenters in the 3-D

case (Figs.

8c and 9c) when compared with the

minimum 1-D solution (Fig.

8d). This tendency

towards shallower depths is also evident when

comparing the minimum 1-D locations with sta-

tion corrections with the initial event locations

(e.g., Fig.

9a, b). Also, the large positive station

(13)

residuals calculated in the inversion for stations in the south (Fig.

5d) hint at low velocities in

that area and thus possibly shallower hypocentral depths. A thick sedimentary cover of up to 10 km, which was found in refraction seismic studies of this region (Bohnhoff et al.

2001; Brönner2003),

may be responsible for the locally reduced data quality due to scattering and subsequent lower

SNR due to damping. The velocity model re- sulting from the VELEST inversion is generally slower than the initial model derived from the onshore network by Meier et al. (2004a) (Fig.

4)

as expected from the fact that the lower velocities in the upper layers in the southern offshore region must somehow be incorporated in an average 1-D velocity model.

Fig. 8 Cross-sections showing the hypocenter distribution along profile AB as obtained by the different localization approaches. a Initial event localization using a velocity model from literature and a gradient method for event localization. b Minimum 1-D velocity model and corresponding station corrections using the gradient method for event localization. c 3-D velocity model and probabilistic event localization. d Minimum 1-D velocity model and corresponding station corrections using the probabilistic event localization. The width of the selected area for projection is 15 km towards both sides of the profile indicated in Fig.7

0 50 100 150 200 250 300

0 20 40

Depth [km] 60

A B

0 50 100 150 200 250 300

0 20 40 Depth [km] 60

A B

0 50 100 150 200 250 300

0 20 40 Depth [km] 60

A B

c)

b) a)

0 50 100 150 200 250 300

0 20 40

Depth [km] 60

A B

d)

^Pliny ^Ptolemy_Messara

(14)

Fig. 9 Cross-sections showing the hypocenter distribution along profile CD as obtained by the different localization approaches. Same model order as in Fig.8

0 50 100 150 200 250 300

0 20 40

Depth [km] 60

0 50 100 150 200 250 300

0 20 40

Depth [km] 60

0 50 100 150 200 250 300

0 20 40

Depth [km] 60

0 50 100 150 200 250 300

0 20 40

Depth [km] 60

a)

b)

c)

d) C

C C

C

D

Pliny Ptolemy D

Strabo

Although the general event distribution is the same for all localization approaches, quantifi- able differences between them do exist. As ex- pected, the minimum 1-D velocity model with corresponding station corrections in conjunction with a gradient method for event localization per- forms significantly better than the starting ve- locity model using the same localization method (Fig.

10a, b). The computed average rms misfit for

the 435 events used in the inversion is 0.26 s when using the minimum 1-D velocity model, a

vp/vs

ra- tio of 1.78 and corresponding station corrections.

This is a significant improvement compared with the average misfit of 0.35 s obtained with the initial

velocity model. Trying to solve for independent S-velocity models does not significantly improve this misfit. Due to the fact that a rather com- plicated model with sharp changes in the

vp/vs

ratio is needed to obtain a similar misfit, 0.24 s compared to the 0.26 s for a constant

vp/vs

ratio, this velocity model with independent S-velocities is rejected.

Despite the incorporation of low velocities and

large

vp/vs

ratios in the southern area, the 3-D

velocity model does not perform better than the

minimum 1-D velocity model with corresponding

station corrections when applying a probabilis-

tic localization scheme to both velocity models,

(15)

although there are several events with very small rms values for the 3-D model. This might be due to the fact that the 3-D velocity model is too simple for the complicated geometry in this curved subduction zone with its local structural heterogeneities. Furthermore, the strong lateral velocity changes make the localization procedure more sensitive to small changes in hypocentral parameters. This effect can be seen in the larger scatter of the hypocenters when compared to the minimum 1-D solution (Fig.

9c, d). Furthermore,

the maximum axis of the error ellipsoid tends to be larger than in the 1-D case (Fig.

10c, d).

Event locations obtained with the minimum 1-D velocity model and a probabilistic approach also show a smaller scatter in the hypocenter dis- tribution (Fig.

9d) than those obtained by a gra-

dient method (Fig.

9b). We performed a synthetic

test to quantify this visual observation. Synthetic travel times were calculated for an event roughly at the center of the offshore network for 11 sta- tions of the amphibian network surrounding this

0 0.2 0.4 0.6 0.8 1

0 50 100 150

rms value

No. of events

0 2 4 6 8 10 12 14 16 18 20

0 100 200 300 400 500

largest axis of error ellipsoid [km]

No. of events

0 2 4 6 8 10 12 14 16 18 20

0 100 200 300 400 500

No. of events

largest axis of error ellipsoid [km]

0 0.2 0.4 0.6 0.8 1

0 50 100 150

rms value

No. of events

a) b)

d) c)

Fig. 10 a Initial velocity model rms misfit. b Final 1-D minimum velocity model rms misfit. The events are the 435 earthquakes used for the VELEST inversion. c Dis- tribution of the largest error ellipsoid half axis for the probabilistic earthquake localization with the simple 3-D

model. d Distribution of the largest error ellipsoid half axis for the probabilistic earthquake localization using the minimum 1-D velocity model with corresponding station corrections in probabilistic earthquake localization mode

(16)

hypocenter. These synthetic travel times were per- turbed with random noise. The added travel time error was normally distributed with a standard deviation of 0.3 s for the P-wave travel time and 0.6 s for the S-wave travel time. Then, the event was located using these perturbed travel times.

This procedure was done 200 times for the linear gradient approach as well as for the probabilistic approach. While both localization schemes show comparable performances with respect to the hor- izontal directions, the probabilistic approach is able to better constrain the depth of the event.

Figure

11

shows the result of the test with re- spect to the recovered depth. The probabilistic approach (Fig.

11b) exhibits a standard deviation

of 1.25 km for the 200 location runs compared to the 1.53 km obtained for the gradient method (Fig.

11a) and avoids termination in local minima.

This contrasts with the result for the gradient method which exhibits a pronounced local maxi- mum at a velocity step at about 34 km depth. Lo- cations obtained by applying a probabilistic event localization scheme with the minimum 1-D veloc-

ity model and corresponding station corrections seem to be on average better constrained than those obtained by the other approaches. Thus, the following discussion of the microseismicity distri- bution is based on the results from the minimum 1-D model with corresponding station corrections used in the probabilistic localization scheme.

Location accuracy for well-resolved events recorded by the network is of the order of less than 5 km for horizontal directions and approx- imately worse by a factor of two in the vertical direction. An idea of the expected error ellip- soids of events recorded by the network can be gained from Fig.

12. In this figure a subset of

the final earthquake catalog is plotted with error ellipsoids obtained by mapping the probability density function for the minimum 1-D model in probabilistic earthquake localization mode. The chosen subset is from a single deployment phase which guarantees that the station configuration does not change. A time period exhibiting activity throughout the whole study region was chosen.

Obviously, the vertical error is considerably larger

24 26 28 30 32 34 36

0 5 10 15 20 25 30 35

Depth [km]

Number of events

24 26 28 30 32 34 36

0 5 10 15 20 25 30 35

Number of events

Depth [km]

a) b)

Fig. 11 Results of synthetic test with linear gradient and probabilistic earthquake location methods. Synthetic travel times to 11 stations of the network surrounding the epi- center were calculated for an event at 34.5° N, 25.75° E, and 30 km depth. Normally distributed random noise with a standard deviation of 0.3 s for P-waves and 0.6 s for

S-waves was added to the synthetic times. The plot shows the results for 200 location runs. a Distribution of depths calculated with the linear gradient approach. Standard deviation 1.53 km. b Distribution of depths calculated with the probabilistic location method. Standard deviation 1.25 km

(17)

24 24.5 25 25.5 26 26.5 27 33.6

33.8 34 34.2 34.4 34.6 34.8 35 35.2 35.4 35.6

Latitude

24 24.5 25 25.5 26 26.5 27

0

50

Longitude

Depth [km]

0 50

33.6 33.8 34 34.2 34.4 34.6 34.8 35 35.2 35.4 35.6

Depth [km]

Strabo Pliny

Fig. 12 Example of error ellipsoids calculated with the minimum 1-D velocity model and the probabilistic localization scheme of NonLinLoc. Plotted are all events with

an rms misfit of less than 0.5 s, a maximum error ellipsoid half axis of no more than 10 km and a magnitude larger than ML 2.0 in the time interval 10/03/04 until 15/04/04

than the horizontal one for less well-constrained events and those at the edges or outside the am- phibian network. A general increase of the error ellipsoid is observed for events locating towards the southern and eastern edges of the amphibian network. The larger error ellipsoids observed for events locating in these areas might be explained by a worse fit of the velocity model to the true situation and a less dense station coverage. In addition, the data quality of the southern stations is not as good as that of the northern ones making proper phase identification and picking more dif- ficult. The horizontal location accuracy however remains high for events in the region of the Pliny trench as indicated by well-constrained events in that region allowing a clear localization of the

seismogenic zones, while it is getting increasingly worse towards the south in the region of the Strabo trench (Fig.

12).

4 Discussion

The microseismic activity in the forearc of the

HSZ south of central and eastern Crete occurring

between July 2003 and June 2004 was studied by

means of an amphibian network covering the en-

tire transtensional on- and offshore graben system

in that area. In the following, the most reliable

hypocenter locations obtained with the proba-

bilistic event localization and using the minimum

1-D velocity model with corresponding station

(18)

18.07.03 16.09.03 15.11.03 14.01.04 14.03.04 13.05.04 0

10 20 30 40 50 60 70 80 90

Date

Events per day Magnitude [Ml]

1

0 2 3 4 5

1.8 2 2.2 2.4 2.6

Mcomp

24 24.5 25 25.5 26 26.5 27

33.6 33.8 34 34.2 34.4 34.6 34.8 35 35.2 35.4 35.6

2003.5431 to 2004.4576

Longitude [deg]

Latitude [deg]

2.5 3 3.5 4 4.5 5

a−^value

24 24.5 25 25.5 26 26.5 27

33.6 33.8 34 34.2 34.4 34.6 34.8 35 35.2 35.4 35.6

2003.5431 to 2004.4576

Longitude [deg]

Latitude [deg]

0.6 0.8 1 1.2 1.4

b−value

24 24.5 25 25.5 26 26.5 27

33.6 33.8 34 34.2 34.4 34.6 34.8 35 35.2 35.4 35.6

2003.5431 to 2004.4576

Longitude [deg]

Latitude [deg]

a) b)

c) d)

Fig. 13 a Magnitude over time plot for all located events during the five recording periods (red stars). Discontinuity of the data set is caused by recovery and redeployment of the OBS stations. Blue bars at the bottom indicate the number of events per day. b Magnitude completeness map of the study region produced using the zmap software package (Wiemer2001) showing the increase in complete-

ness magnitude with increasing distance from the island of Crete. White regions indicate areas with insufficient data for the calculation of the completeness magnitude. c Calculated b values for the LIBNET data set in the study region. d Calculated a values for the LIBNET data set in the study region

corrections are used to interpret the microseismic- ity distribution.

4.1 Microseismicity offshore of Crete

Prominent crustal microseismic activity within the Aegean plate in the area of the Ptolemy trench

was observed throughout the whole observation period (Fig.

13a). This activity close to the south-

ern coast of Crete was also identified in earlier microseismicity studies of the region (Delibasis et al.

1999; Meier et al. 2004b; Becker et al.

2006) using land-based seismic networks. Combin-

ing our findings with the ISC catalog (Engdahl

et al.

1998) which also lists seismic events with

(19)

intermediate magnitudes in that area, we conclude that the Ptolemy structure constitutes a contin- uously seismically active feature in the Aegean crust. This intracrustal activity extends from near surface depths down to the presumed plate inter- face at approximately 40 km depth and was traced by this study in west–east direction from offshore Messara in the west (Fig.

8) up to the eastern

shoreline of Crete and possibly even further east towards the island of Karpathos (Fig.

7). In a

study emphasizing basin-scale structural relations, ten Veen and Kleinspehn (2003) suggest an active transtensional regime on- and offshore central and eastern Crete resulting in forearc slivers separated from the Cretan block due to internal deformation of the Aegean crust in that area. They arrive at this conclusion by analyzing motions on 070

^◦

sinis- tral faults in the Messara region. The development of these slivers is explained by the blocking of the further outward expansion of the Aegean plate boundary in western Crete due to the incipient collision with Africa and the simultaneous expan- sion of the Aegean crust by the stretching of the Cretan–Rhodes forearc in the east caused by the still active slab rollback in that area. Although the findings of this study have been questioned (Ring and Borchert

2003), seismic studies show

similar wedge-shaped sedimentary basins in the vicinity of the Ptolemy and Pliny trench (Bohn- hoff et al.

2001), suggesting a similar origin of

the two structures. Fault plane solutions (Fig.

1)

and analysis of GPS measurements (Kahle et al.

2000) indicate sinistral motions in the area of

the Pliny and Strabo trenches. The presence of microearthquake clusters found in earlier studies (Becker et al.

2006) suggests that the Ptolemy

trench represents a zone of crustal weakness al- lowing the ascent of fluids from the plate contact.

Towards the south between the Ptolemy and the Pliny trenches, we observe a sharp drop of crustal seismicity within the overriding Aegean plate and seismicity is nearly exclusively found at the presumed plate interface (Figs.

8

and

9).

The absence of intraplate microseismic activity between the Ptolemy and Pliny trenches is in con- trast to the rather strong microseismic activity ob- served further south between the Pliny and Strabo trenches. This absence of microseismic activity might indicate that no or only very little internal

deformation is occurring at this part of the upper plate. Therefore, the microseismicity distribution found in this study suggests that the Ptolemy and Pliny trenches are the borders of a forearc sliver separated from the rest of the Aegean forearc.